Skip to content

Google's Stitch Loop watches live products and proposes code fixes

News
A magnifying glass on a metal stand examines the smooth silver lid of a closed laptop on a blue workbench.
AI-generated editorial illustration of continuous product inspection.
Google has opened a waitlist for Stitch Loop. It aims to help teams fix apps after launch. The AI tool reads code and live data, then suggests changes.
The waitlist opened worldwide on November 18. Google AI Ultra users who meet its rules get first access. Google calls Stitch Loop an experiment.

How Stitch Loop turns data into code

Teams link GitHub and data such as logs and feedback. Stitch Loop uses that context to find work worth doing. Its first goals cover security, speed and accessibility.
Teams can add a goal of their own. One example is to help more shoppers finish a purchase.
In Google's launch post, each proposed fix has a scope and tradeoffs. Once a team picks a path, Loop writes the code. It does that work in a cloud sandbox.
It then opens a pull request for review. A person must review the change before it merges.

CodeMender helps check the code

The security loop uses Gemini 4 Carbon and CodeMender. Our Carbon report covers the model's work with other agents.
DeepMind's CodeMender research describes tools that find the cause of a flaw. They also check a fix for errors it could add.
The tools can inspect code and test how it runs. They can compare the old code with the new version. The research team had people review patches before sending them upstream.

Keep track of what changes

Loop tracks live data once a fix goes out. It uses the result to guide its next suggestion. When a team meets a goal, it keeps watching that measure.
Google says other coding agents can use Loop's context. They connect through its command-line tools or Model Context Protocol. Teams can also ask for insights through Slack.
The work brings code and product data into one cloud flow. Our local and cloud AI guide explains where those tools run.
loading

Loading